Panicle-Cloud: An Open and AI-Powered Cloud Computing Platform for Quantifying Rice Panicles from Drone-Collected...

Area of Science:

• Agricultural Science
• Plant Science
• Computational Biology

Background:

• Rice (Oryza sativa) is a global staple food, necessitating yield improvements amidst climate change.
• Accurate phenotyping of yield-related traits like panicle number per unit area (PNpM²) is vital for rice breeding.
• Current methods for large-scale panicle quantification are challenging due to field complexity and trait variability.

Purpose of the Study:

• To develop an AI-powered platform for automated rice panicle quantification from drone imagery.
• To create an open, diverse dataset for training and validating AI models for rice panicle detection.
• To provide a user-friendly toolkit for rice breeders to assess yield potential.

Main Methods:

• Development of Panicle-Cloud, an AI-driven cloud platform for analyzing drone-collected rice imagery.
• Creation of a diverse, expert-annotated rice panicle detection dataset.
• Integration and application of deep learning models, including Panicle-AI, for panicle quantification.
• Field trials across different growth stages and seasons to optimize image acquisition and model performance.

Main Results:

• Panicle-Cloud reliably quantifies the panicle number per unit area (PNpM²) trait.
• AI models demonstrated high accuracy in classifying yield production based on PNpM².
• Optimized image resolutions and timing for effective field phenotyping were identified.

Conclusions:

• Panicle-Cloud offers a significant advancement in high-throughput phenotyping of rice panicles.
• The platform provides a valuable, reliable toolkit for breeders to screen and select superior rice varieties.
• This AI-driven approach supports efficient breeding programs to enhance rice yield potential.